Temporal-spatial Feature Learning of DCE-MR Images via 3DCNN

Code for paper:

Temporal-spatial Feature Learning of Dynamic Contrast Enhanced-MR Images via 3D Convolutional Neural Networks

Noninvasive Evaluation of the Pathologic Grade of Hepatocellular Carcinoma Using MCF-3DCNN: A Pilot Study

Requirements

Python 2.7

TensorFlow == 1.4.0

Keras == 2.2.4
For keras2.0.0 compatibility checkout tag keras2.0.0

Run:

1. Start the training using:

python main.py -c configs/fusion_config.json  # MCF-3D-CNN

python main.py -c configs/3dcnn_config.json   # 3DCNN

2. Start Tensorboard visualization using:

tensorboard --logdir=experiments/Year-Month-Day/Ex-name/logs

Data

The proprietary of the data belongs to Beijing Friendship Hospital. You can get access to anonymous data here.

Tensor-based data representation

MCF-3DCNN architecture

Results

Feature maps of C1 and C2 convolution layer

One Vs. Other

A multi-classification problem is transformed into multiple binary classification problems. The results are as follow:

The average area under the ROC curve for 3DCNN for discriminating poorly, moderately and well differentiated HCCs.

Reference

Keras-Project-Template

Citation

If you use this code or data for your research, please cite our papers.

@inproceedings{IGTA 2018,    
    title={Temporal-Spatial Feature Learning of Dynamic Contrast Enhanced-MR Images via 3D Convolutional Neural Networks},    
    author={Jia X., Xiao Y., Yang D., Yang Z., Wang X., Liu Y},    
    booktitle={Image and Graphics Technologies and Applications. IGTA 2018. Communications in Computer and Information Science},    
    year={2018}    
}

Da-wei Yang, Xi-bin Jia, Yu-jie Xiao, Xiao-pei Wang, Zhen-chang Wang, and Zheng-han Yang, “Noninvasive Evaluation of the Pathologic Grade of Hepatocellular Carcinoma Using MCF-3DCNN: A Pilot Study,” BioMed Research International, vol. 2019, Article ID 9783106, 12 pages, 2019. https://doi.org/10.1155/2019/9783106.